Too Little Storage Stops the Expensive End
Reblend, Scheduling, and In-Process Inventory
A CPG manufacturer found its in-process storage requirement was not a fixed number. It moved with scheduling rules and with how much material was being recycled back through the process. Get it wrong in the tight direction and the consequence was not a slow line — it was shutting down expensive upstream processing.
Why reblend makes storage hard to size
Reblend is material re-introduced into the process rather than discarded — ordinary practice, and good economics. It also makes in-process inventory much harder to predict, because the flow into storage is no longer just the making rate. It is the making rate plus whatever is being recycled, and the recycle stream is itself a function of what has been running and how it has been scheduled.
That creates a genuine circularity: scheduling changes the reblend volume, reblend volume changes the storage requirement, and the storage requirement constrains what can be scheduled. Sizing that with a static calculation means picking one operating point and hoping the plant stays near it.
The asymmetry matters. Oversized storage costs capital and floor space. Undersized storage backs up into the making process and stops it — and upstream processing is the expensive end of this plant. The two errors are not equally forgivable, which is exactly when you want a distribution rather than a point estimate.
What the model showed
The team used discrete-rate simulation to study how in-process inventory actually behaved over time, and how scheduling policy reshaped the utilisation problem rather than merely shifting it. This is the bulk-flow, rate-based case the method was designed for: continuous material, recirculation, and a storage buffer whose adequacy depends on dynamics rather than averages.
The outcome was a set of better scheduling strategies, developed and validated against the model before being put into practice.
The part that says the most
Outcome
| Measure | Value |
|---|---|
| Scheduling strategies developed and validated | before deployment |
| Sites the model was re-implemented at | 12+ similar factories |
| In-process inventory management | materially improved |
A model that solves one plant is a consulting deliverable. A model re-implemented at a dozen similar factories is something else: evidence that the behaviour it captured was structural rather than local. Once a rate-based problem of this kind is properly solved, the solution usually transfers to operations built the same way — which is the strongest argument for modelling the mechanism rather than fitting the history of one site.
On the evidence
This account comes from our own project record rather than a published paper. The client is not named in the source and is not named here. The replication count is reported as it appears in that record.
Related: How big should a buffer be? · Starved vs Blocked · All case studies